Resources

Agent resources

Choose models, audit local usage and read the docs. Discovery and model advice need no account.

Agent onboarding

/agents.md tells an agent how to find a key, configure the base URL, send a real request, and recover from common errors. People setting up an agent can follow Onboard an agent.

Retrieve one page

Append .md to a docs or integration URL. Agents that send Accept: text/markdown can use the normal URL.

Two ways to fetch the models guide
curl https://minirouter.sh/docs/models.md

curl -H "Accept: text/markdown" \
  https://minirouter.sh/docs/models

Index or full corpus

Use the short index when the agent can fetch individual pages. Use the full file when a tool needs all minirouter documentation in one request.

Machine-readable documentation
# Short index: fetch only the pages you need
https://minirouter.sh/llms.txt

# Complete docs and integration guides
https://minirouter.sh/llms-full.txt

Public MCP server

Connect a Streamable HTTP MCP client to the public endpoint. It can search the docs, fetch a page and recommend model-and-effort configurations. All operations are public and read-only. Never send credentials, prompts or usage files to these tools.

MCP connection
https://minirouter.sh/mcp

Tools: search_docs, get_doc, recommend_models
Resources: every docs and integration Markdown page

Server metadata: /.well-known/mcp/server-card.json

Agent skills

  • audit spend: Review local usage without uploading your file.
  • choose model: Compare models for your task and token mix.
  • evaluate switch: Compare task quality and cost within an approved budget.
  • optimize usage: Find alternatives using your account’s settled usage.

Download and verify each skill before running it. Connected analysis uses an existing key; evaluations require approved spending.

The reusable skill lives at /skills/minirouter/SKILL.md. Compatible tools can discover it from /.well-known/agent-skills/index.json.

Model advice

Public evidence snapshot supplies current capabilities, prices and attributed benchmarks. Use recommend_models to compare a task and token mix. Estimates hold token counts constant; test candidates on your own tasks before switching.

Coding-agent setup

Setup has complete configuration blocks for Cline, Kilo Code, OpenCode, Continue, LiteLLM and SillyTavern. The integration guides add recommended models and troubleshooting.